Install
Act as an active guide through Context Gathering, Refinement and Reader TestingWho it's for
doc-coauthoring is for people who need structured text: documents, proposals, technical specs, decision memos, or similar. It is not an auto-delivery tool. It actively walks you through three stages: Context Gathering (gather context), Refinement & Structure (sharpen and structure), Reader Testing (reader testing).
The trigger is written plainly: use it when you mention writing a document, making a proposal, drafting a spec, or a similar task. If you only want a casual chat and will not accept staged guidance, the guide lets you refuse this structure and switch to freeform; once you accept, advance by stage instead of jumping to a finished draft.
It does not fit being an installer. The guide gives no install command and does not say that placing it in some directory makes it usable. If the current assistant cannot yet work in these three stages, the existing guide stops here; do not reverse-engineer an install path from the page address. It also does not fit sending a first draft to others without reader testing.
First try in 3 minutes
The official first path is to start co-writing, not to install. The guide page is
https://raw.githubusercontent.com/anthropics/skills/main/skills/doc-coauthoring/SKILL.md
Follow only the stages on that page. Do not write an install phrase that never appears.
Prerequisite: the current assistant can already guide you through doc-coauthoring steps, and you have a low-risk topic on hand, for example a one-page internal decision-memo draft. There is no install command, so the “location” is this conversation, not a guessed skill directory.
Low-risk first task: say clearly “I want to write a decision memo,” accept the three-stage guidance, but finish only Context Gathering. Answer the assistant's questions about readers, purpose, and known materials first; do not rush a full text, and do not send it outward. If the assistant mentions Slack, Teams, Google Drive, SharePoint, or other MCP that could pull context directly, first confirm those integrations are actually available on your side; if not, switch to paste or spoken notes — do not pretend they are already connected.
Success signal: the assistant states it will walk you through the three stages and starts gathering context, instead of dumping a complete document. Before the gathering stage ends, do not approve large writes to files. The guide does not say which editor you must use, and does not name a file format.
Use cases
Case 1: Co-write a spec or proposal in three stages. Start when you mention writing a document, making a proposal, or drafting a spec. The assistant acts as an active guide: first gather context, then sharpen structure and wording. This fits when you already have material but need someone to chase gaps by stage. The guide does not give a question list per stage, so you can only “accept the guidance and confirm section by section.” If the assistant skips gathering and jumps to a finished draft, ask to return to Context Gathering. The guide page is still
https://raw.githubusercontent.com/anthropics/skills/main/skills/doc-coauthoring/SKILL.md
Case 2: Do reader testing before the draft is done. Stage three is to have a fresh Claude with no prior context read this document, catch blind spots, then let other people read it. This fits when your own writing feels too smooth and you worry outsiders will not follow. The guide does not write how to open a new session in one click, and does not write a test-question template; so you can only “open another no-context session, paste the document, and see whether the other side can restate the points.” Until that test is done, do not treat the document as ready. If integrations are available they may pull context, but they do not replace reader testing.
Boundaries
No install command: the guide only defines the skill identity and the three-stage flow; do not treat the page path as an installer. Trigger boundary: for writing documents, proposals, specs, or similar structured content; other writing tasks are not named. Flow boundary: the three stages are gather context, sharpen and structure, and reader testing; reader testing must use a fresh Claude with no prior context. Integration boundary: Slack, Teams, Google Drive, SharePoint, or other MCP may pull context only when “available”; the guide does not guarantee those integrations exist, and does not write enablement steps. If you refuse the guidance you may switch to freeform, but do not mistake freeform for having finished reader testing.